feat(server): fully accelerated nvenc (#9452)

* use arrayContaining * libplacebo for nvenc update dockerfile * tweaks * update nvenc options * tweak settings * refactor * toggle for hardware decoding, software / hardware decoding for nvenc and rkmpp * fix software tone-mapping not being applied * separate configs for hw/sw * update api * add hw decode toggle * fix mutating config * remove `version` flag * fix config type * remove submodule * handle temporal AQ * remove duplicate tests * use `tonemap_opencl` * wording * update docs
2024-09-21 10:37:20 +00:00 · 2024-05-16 13:30:26 -04:00 · 2024-05-16 13:30:26 -04:00 · d8eca168ca
commit d8eca168ca
parent 64636c0618
15 changed files with 255 additions and 64 deletions
--- a/docker/hwaccel.ml.yml
+++ b/docker/hwaccel.ml.yml
@ -1,9 +1,7 @@
-version: "3.8"
-
 # Configurations for hardware-accelerated machine learning

 # If using Unraid or another platform that doesn't allow multiple Compose files,
-# you can inline the config for a backend by copying its contents 
+# you can inline the config for a backend by copying its contents
 # into the immich-machine-learning service in the docker-compose.yml file.

 # See https://immich.app/docs/features/ml-hardware-acceleration for info on usage.
@ -30,7 +28,7 @@ services:

  openvino:
    device_cgroup_rules:
-      - "c 189:* rmw"
+      - 'c 189:* rmw'
    devices:
      - /dev/dri:/dev/dri
    volumes:
--- a/docker/hwaccel.transcoding.yml
+++ b/docker/hwaccel.transcoding.yml
@ -1,5 +1,3 @@
-version: "3.8"
-
 # Configurations for hardware-accelerated transcoding

 # If using Unraid or another platform that doesn't allow multiple Compose files,
--- a/docs/docs/features/hardware-transcoding.md
+++ b/docs/docs/features/hardware-transcoding.md
@ -22,7 +22,8 @@ You do not need to redo any transcoding jobs after enabling hardware acceleratio
 - WSL2 does not support Quick Sync.
 - Raspberry Pi is currently not supported.
 - Two-pass mode is only supported for NVENC. Other APIs will ignore this setting.
- Only encoding is currently hardware accelerated, so the CPU is still used for software decoding and tone-mapping.
+- By default, only encoding is currently hardware accelerated. This means the CPU is still used for software decoding and tone-mapping.
+  - NVENC and RKMPP can be fully accelerated by enabling hardware decoding in the video transcoding settings.
 - Hardware dependent
  - Codec support varies, but H.264 and HEVC are usually supported.
    - Notably, NVIDIA and AMD GPUs do not support VP9 encoding.
@ -65,6 +66,7 @@ For RKMPP to work:

 3. Redeploy the `immich-microservices` container with these updated settings.
 4. In the Admin page under `Video transcoding settings`, change the hardware acceleration setting to the appropriate option and save.
+5. (Optional) If using a compatible backend, you may enable hardware decoding for optimal performance.

 #### Single Compose File

--- a/mobile/openapi/doc/SystemConfigFFmpegDto.md
+++ b/mobile/openapi/doc/SystemConfigFFmpegDto.md
@ -9,6 +9,7 @@ import 'package:openapi/api.dart';
 Name | Type | Description | Notes
 ------------ | ------------- | ------------- | -------------
 **accel** | [**TranscodeHWAccel**](TranscodeHWAccel.md) |  | 
+**accelDecode** | **bool** |  | 
 **acceptedAudioCodecs** | [**List<AudioCodec>**](AudioCodec.md) |  | [default to const []]
 **acceptedVideoCodecs** | [**List<VideoCodec>**](VideoCodec.md) |  | [default to const []]
 **bframes** | **int** |  | 
--- a/mobile/openapi/lib/model/system_config_f_fmpeg_dto.dart
+++ b/mobile/openapi/lib/model/system_config_f_fmpeg_dto.dart
@ -14,6 +14,7 @@ class SystemConfigFFmpegDto {
  /// Returns a new [SystemConfigFFmpegDto] instance.
  SystemConfigFFmpegDto({
    required this.accel,
+    required this.accelDecode,
    this.acceptedAudioCodecs = const [],
    this.acceptedVideoCodecs = const [],
    required this.bframes,
@ -37,6 +38,8 @@ class SystemConfigFFmpegDto {

  TranscodeHWAccel accel;

+  bool accelDecode;
+
  List<AudioCodec> acceptedAudioCodecs;

  List<VideoCodec> acceptedVideoCodecs;
@ -87,6 +90,7 @@ class SystemConfigFFmpegDto {
  @override
  bool operator ==(Object other) => identical(this, other) || other is SystemConfigFFmpegDto &&
    other.accel == accel &&
+    other.accelDecode == accelDecode &&
    _deepEquality.equals(other.acceptedAudioCodecs, acceptedAudioCodecs) &&
    _deepEquality.equals(other.acceptedVideoCodecs, acceptedVideoCodecs) &&
    other.bframes == bframes &&
@ -111,6 +115,7 @@ class SystemConfigFFmpegDto {
  int get hashCode =>
    // ignore: unnecessary_parenthesis
    (accel.hashCode) +
+    (accelDecode.hashCode) +
    (acceptedAudioCodecs.hashCode) +
    (acceptedVideoCodecs.hashCode) +
    (bframes.hashCode) +
@ -132,11 +137,12 @@ class SystemConfigFFmpegDto {
    (twoPass.hashCode);

  @override
-  String toString() => 'SystemConfigFFmpegDto[accel=$accel, acceptedAudioCodecs=$acceptedAudioCodecs, acceptedVideoCodecs=$acceptedVideoCodecs, bframes=$bframes, cqMode=$cqMode, crf=$crf, gopSize=$gopSize, maxBitrate=$maxBitrate, npl=$npl, preferredHwDevice=$preferredHwDevice, preset=$preset, refs=$refs, targetAudioCodec=$targetAudioCodec, targetResolution=$targetResolution, targetVideoCodec=$targetVideoCodec, temporalAQ=$temporalAQ, threads=$threads, tonemap=$tonemap, transcode=$transcode, twoPass=$twoPass]';
+  String toString() => 'SystemConfigFFmpegDto[accel=$accel, accelDecode=$accelDecode, acceptedAudioCodecs=$acceptedAudioCodecs, acceptedVideoCodecs=$acceptedVideoCodecs, bframes=$bframes, cqMode=$cqMode, crf=$crf, gopSize=$gopSize, maxBitrate=$maxBitrate, npl=$npl, preferredHwDevice=$preferredHwDevice, preset=$preset, refs=$refs, targetAudioCodec=$targetAudioCodec, targetResolution=$targetResolution, targetVideoCodec=$targetVideoCodec, temporalAQ=$temporalAQ, threads=$threads, tonemap=$tonemap, transcode=$transcode, twoPass=$twoPass]';

  Map<String, dynamic> toJson() {
    final json = <String, dynamic>{};
      json[r'accel'] = this.accel;
+      json[r'accelDecode'] = this.accelDecode;
      json[r'acceptedAudioCodecs'] = this.acceptedAudioCodecs;
      json[r'acceptedVideoCodecs'] = this.acceptedVideoCodecs;
      json[r'bframes'] = this.bframes;
@ -168,6 +174,7 @@ class SystemConfigFFmpegDto {

      return SystemConfigFFmpegDto(
        accel: TranscodeHWAccel.fromJson(json[r'accel'])!,
+        accelDecode: mapValueOfType<bool>(json, r'accelDecode')!,
        acceptedAudioCodecs: AudioCodec.listFromJson(json[r'acceptedAudioCodecs']),
        acceptedVideoCodecs: VideoCodec.listFromJson(json[r'acceptedVideoCodecs']),
        bframes: mapValueOfType<int>(json, r'bframes')!,
@ -235,6 +242,7 @@ class SystemConfigFFmpegDto {
  /// The list of required keys that must be present in a JSON.
  static const requiredKeys = <String>{
    'accel',
+    'accelDecode',
    'acceptedAudioCodecs',
    'acceptedVideoCodecs',
    'bframes',
--- a/mobile/openapi/test/system_config_f_fmpeg_dto_test.dart
+++ b/mobile/openapi/test/system_config_f_fmpeg_dto_test.dart
@ -21,6 +21,11 @@ void main() {
      // TODO
    });

+    // bool accelDecode
+    test('to test the property `accelDecode`', () async {
+      // TODO
+    });
+
    // List<AudioCodec> acceptedAudioCodecs (default value: const [])
    test('to test the property `acceptedAudioCodecs`', () async {
      // TODO
--- a/open-api/immich-openapi-specs.json
+++ b/open-api/immich-openapi-specs.json
@ -10050,6 +10050,9 @@
          "accel": {
            "$ref": "#/components/schemas/TranscodeHWAccel"
          },
+          "accelDecode": {
+            "type": "boolean"
+          },
          "acceptedAudioCodecs": {
            "items": {
              "$ref": "#/components/schemas/AudioCodec"
@ -10125,6 +10128,7 @@
        },
        "required": [
          "accel",
+          "accelDecode",
          "acceptedAudioCodecs",
          "acceptedVideoCodecs",
          "bframes",
--- a/open-api/typescript-sdk/src/fetch-client.ts
+++ b/open-api/typescript-sdk/src/fetch-client.ts
@ -863,6 +863,7 @@ export type AssetFullSyncDto = {
 };
 export type SystemConfigFFmpegDto = {
    accel: TranscodeHWAccel;
+    accelDecode: boolean;
    acceptedAudioCodecs: AudioCodec[];
    acceptedVideoCodecs: VideoCodec[];
    bframes: number;
--- a/server/src/config.ts
+++ b/server/src/config.ts
@ -97,6 +97,7 @@ export interface SystemConfig {
    preferredHwDevice: string;
    transcode: TranscodePolicy;
    accel: TranscodeHWAccel;
+    accelDecode: boolean;
    tonemap: ToneMapping;
  };
  job: Record<ConcurrentQueueName, { concurrency: number }>;
@ -228,6 +229,7 @@ export const defaults = Object.freeze<SystemConfig>({
    transcode: TranscodePolicy.REQUIRED,
    tonemap: ToneMapping.HABLE,
    accel: TranscodeHWAccel.DISABLED,
+    accelDecode: false,
  },
  job: {
    [QueueName.BACKGROUND_TASK]: { concurrency: 5 },
--- a/server/src/dtos/system-config.dto.ts
+++ b/server/src/dtos/system-config.dto.ts
@ -132,6 +132,9 @@ export class SystemConfigFFmpegDto {
  @ApiProperty({ enumName: 'TranscodeHWAccel', enum: TranscodeHWAccel })
  accel!: TranscodeHWAccel;

+  @ValidateBoolean()
+  accelDecode!: boolean;
+
  @IsEnum(ToneMapping)
  @ApiProperty({ enumName: 'ToneMapping', enum: ToneMapping })
  tonemap!: ToneMapping;
--- a/server/src/services/media.service.spec.ts
+++ b/server/src/services/media.service.spec.ts
@ -1335,6 +1335,51 @@ describe(MediaService.name, () => {
      );
    });

+    it('should use hardware decoding for nvenc if enabled', async () => {
+      mediaMock.probe.mockResolvedValue(probeStub.matroskaContainer);
+      systemMock.get.mockResolvedValue({
+        ffmpeg: { accel: TranscodeHWAccel.NVENC, accelDecode: true },
+      });
+      assetMock.getByIds.mockResolvedValue([assetStub.video]);
+      await sut.handleVideoConversion({ id: assetStub.video.id });
+      expect(mediaMock.transcode).toHaveBeenCalledWith(
+        '/original/path.ext',
+        'upload/encoded-video/user-id/as/se/asset-id.mp4',
+        {
+          inputOptions: expect.arrayContaining([
+            '-hwaccel cuda',
+            '-hwaccel_output_format cuda',
+            '-noautorotate',
+            '-threads 1',
+          ]),
+          outputOptions: expect.arrayContaining([expect.stringContaining('scale_cuda=-2:720:format=nv12')]),
+          twoPass: false,
+        },
+      );
+    });
+
+    it('should use hardware tone-mapping for nvenc if hardware decoding is enabled and should tone map', async () => {
+      mediaMock.probe.mockResolvedValue(probeStub.videoStreamHDR);
+      systemMock.get.mockResolvedValue({
+        ffmpeg: { accel: TranscodeHWAccel.NVENC, accelDecode: true },
+      });
+      assetMock.getByIds.mockResolvedValue([assetStub.video]);
+      await sut.handleVideoConversion({ id: assetStub.video.id });
+      expect(mediaMock.transcode).toHaveBeenCalledWith(
+        '/original/path.ext',
+        'upload/encoded-video/user-id/as/se/asset-id.mp4',
+        {
+          inputOptions: expect.arrayContaining(['-hwaccel cuda', '-hwaccel_output_format cuda']),
+          outputOptions: expect.arrayContaining([
+            expect.stringContaining(
+              'tonemap_cuda=desat=0:matrix=bt709:primaries=bt709:range=pc:tonemap=hable:transfer=bt709:format=nv12',
+            ),
+          ]),
+          twoPass: false,
+        },
+      );
+    });
+
    it('should set options for qsv', async () => {
      storageMock.readdir.mockResolvedValue(['renderD128']);
      mediaMock.probe.mockResolvedValue(probeStub.matroskaContainer);
@ -1633,8 +1678,9 @@ describe(MediaService.name, () => {

    it('should set options for rkmpp', async () => {
      storageMock.readdir.mockResolvedValue(['renderD128']);
+      storageMock.stat.mockResolvedValue({ ...new Stats(), isFile: () => true, isCharacterDevice: () => true });
      mediaMock.probe.mockResolvedValue(probeStub.matroskaContainer);
-      systemMock.get.mockResolvedValue({ ffmpeg: { accel: TranscodeHWAccel.RKMPP } });
+      systemMock.get.mockResolvedValue({ ffmpeg: { accel: TranscodeHWAccel.RKMPP, accelDecode: true } });
      assetMock.getByIds.mockResolvedValue([assetStub.video]);
      await sut.handleVideoConversion({ id: assetStub.video.id });
      expect(mediaMock.transcode).toHaveBeenCalledWith(
@ -1663,10 +1709,12 @@ describe(MediaService.name, () => {

    it('should set vbr options for rkmpp when max bitrate is enabled', async () => {
      storageMock.readdir.mockResolvedValue(['renderD128']);
+      storageMock.stat.mockResolvedValue({ ...new Stats(), isFile: () => true, isCharacterDevice: () => true });
      mediaMock.probe.mockResolvedValue(probeStub.videoStreamVp9);
      systemMock.get.mockResolvedValue({
        ffmpeg: {
          accel: TranscodeHWAccel.RKMPP,
+          accelDecode: true,
          maxBitrate: '10000k',
          targetVideoCodec: VideoCodec.HEVC,
        },
@ -1686,9 +1734,10 @@ describe(MediaService.name, () => {

    it('should set cqp options for rkmpp when max bitrate is disabled', async () => {
      storageMock.readdir.mockResolvedValue(['renderD128']);
+      storageMock.stat.mockResolvedValue({ ...new Stats(), isFile: () => true, isCharacterDevice: () => true });
      mediaMock.probe.mockResolvedValue(probeStub.matroskaContainer);
      systemMock.get.mockResolvedValue({
-        ffmpeg: { accel: TranscodeHWAccel.RKMPP, crf: 30, maxBitrate: '0' },
+        ffmpeg: { accel: TranscodeHWAccel.RKMPP, accelDecode: true, crf: 30, maxBitrate: '0' },
      });
      assetMock.getByIds.mockResolvedValue([assetStub.video]);
      await sut.handleVideoConversion({ id: assetStub.video.id });
@ -1707,7 +1756,9 @@ describe(MediaService.name, () => {
      storageMock.readdir.mockResolvedValue(['renderD128']);
      storageMock.stat.mockResolvedValue({ ...new Stats(), isFile: () => true, isCharacterDevice: () => true });
      mediaMock.probe.mockResolvedValue(probeStub.videoStreamHDR);
-      systemMock.get.mockResolvedValue({ ffmpeg: { accel: TranscodeHWAccel.RKMPP, crf: 30, maxBitrate: '0' } });
+      systemMock.get.mockResolvedValue({
+        ffmpeg: { accel: TranscodeHWAccel.RKMPP, accelDecode: true, crf: 30, maxBitrate: '0' },
+      });
      assetMock.getByIds.mockResolvedValue([assetStub.video]);
      await sut.handleVideoConversion({ id: assetStub.video.id });
      expect(mediaMock.transcode).toHaveBeenCalledWith(
@ -1724,6 +1775,54 @@ describe(MediaService.name, () => {
        },
      );
    });
+
+    it('should use software decoding and tone-mapping if hardware decoding is disabled', async () => {
+      storageMock.readdir.mockResolvedValue(['renderD128']);
+      storageMock.stat.mockResolvedValue({ ...new Stats(), isFile: () => true, isCharacterDevice: () => true });
+      mediaMock.probe.mockResolvedValue(probeStub.videoStreamHDR);
+      systemMock.get.mockResolvedValue({
+        ffmpeg: { accel: TranscodeHWAccel.RKMPP, accelDecode: false, crf: 30, maxBitrate: '0' },
+      });
+      assetMock.getByIds.mockResolvedValue([assetStub.video]);
+      await sut.handleVideoConversion({ id: assetStub.video.id });
+      expect(mediaMock.transcode).toHaveBeenCalledWith(
+        '/original/path.ext',
+        'upload/encoded-video/user-id/as/se/asset-id.mp4',
+        {
+          inputOptions: [],
+          outputOptions: expect.arrayContaining([
+            expect.stringContaining(
+              'zscale=t=linear:npl=100,tonemap=hable:desat=0,zscale=p=bt709:t=bt709:m=bt709:range=pc,format=yuv420p',
+            ),
+          ]),
+          twoPass: false,
+        },
+      );
+    });
+
+    it('should use software decoding and tone-mapping if opencl is not available', async () => {
+      storageMock.readdir.mockResolvedValue(['renderD128']);
+      storageMock.stat.mockResolvedValue({ ...new Stats(), isFile: () => false, isCharacterDevice: () => false });
+      mediaMock.probe.mockResolvedValue(probeStub.videoStreamHDR);
+      systemMock.get.mockResolvedValue({
+        ffmpeg: { accel: TranscodeHWAccel.RKMPP, accelDecode: true, crf: 30, maxBitrate: '0' },
+      });
+      assetMock.getByIds.mockResolvedValue([assetStub.video]);
+      await sut.handleVideoConversion({ id: assetStub.video.id });
+      expect(mediaMock.transcode).toHaveBeenCalledWith(
+        '/original/path.ext',
+        'upload/encoded-video/user-id/as/se/asset-id.mp4',
+        {
+          inputOptions: [],
+          outputOptions: expect.arrayContaining([
+            expect.stringContaining(
+              'zscale=t=linear:npl=100,tonemap=hable:desat=0,zscale=p=bt709:t=bt709:m=bt709:range=pc,format=yuv420p',
+            ),
+          ]),
+          twoPass: false,
+        },
+      );
+    });
  });

  it('should tonemap when policy is required and video is hdr', async () => {
--- a/server/src/services/media.service.ts
+++ b/server/src/services/media.service.ts
@ -36,9 +36,11 @@ import {
  AV1Config,
  H264Config,
  HEVCConfig,
-  NVENCConfig,
+  NvencHwDecodeConfig,
+  NvencSwDecodeConfig,
  QSVConfig,
-  RKMPPConfig,
+  RkmppHwDecodeConfig,
+  RkmppSwDecodeConfig,
  ThumbnailConfig,
  VAAPIConfig,
  VP9Config,
@ -360,8 +362,7 @@ export class MediaService {
          `Error occurred during transcoding. Retrying with ${config.accel.toUpperCase()} acceleration disabled.`,
        );
      }
-      config.accel = TranscodeHWAccel.DISABLED;
-      transcodeOptions = await this.getCodecConfig(config).then((c) =>
+      transcodeOptions = await this.getCodecConfig({ ...config, accel: TranscodeHWAccel.DISABLED }).then((c) =>
        c.getOptions(target, mainVideoStream, mainAudioStream),
      );
      await this.mediaRepository.transcode(input, output, transcodeOptions);
@ -494,7 +495,7 @@ export class MediaService {
    let handler: VideoCodecHWConfig;
    switch (config.accel) {
      case TranscodeHWAccel.NVENC: {
-        handler = new NVENCConfig(config);
+        handler = config.accelDecode ? new NvencHwDecodeConfig(config) : new NvencSwDecodeConfig(config);
        break;
      }
      case TranscodeHWAccel.QSV: {
@ -506,7 +507,10 @@ export class MediaService {
        break;
      }
      case TranscodeHWAccel.RKMPP: {
-        handler = new RKMPPConfig(config, await this.getDevices(), await this.hasOpenCL());
+        handler =
+          config.accelDecode && (await this.hasOpenCL())
+            ? new RkmppHwDecodeConfig(config, await this.getDevices())
+            : new RkmppSwDecodeConfig(config, await this.getDevices());
        break;
      }
      default: {
--- a/server/src/services/system-config.service.spec.ts
+++ b/server/src/services/system-config.service.spec.ts
@ -66,6 +66,7 @@ const updatedConfig = Object.freeze<SystemConfig>({
    preferredHwDevice: 'auto',
    transcode: TranscodePolicy.REQUIRED,
    accel: TranscodeHWAccel.DISABLED,
+    accelDecode: false,
    tonemap: ToneMapping.HABLE,
  },
  logging: {
--- a/server/src/utils/media.ts
+++ b/server/src/utils/media.ts
@ -26,14 +26,18 @@ class BaseConfig implements VideoCodecSWConfig {
      }
    }

-    options.outputOptions.push(...this.getPresetOptions(), ...this.getThreadOptions(), ...this.getBitrateOptions());
+    options.outputOptions.push(
+      ...this.getPresetOptions(),
+      ...this.getOutputThreadOptions(),
+      ...this.getBitrateOptions(),
+    );

    return options;
  }

  // eslint-disable-next-line @typescript-eslint/no-unused-vars
  getBaseInputOptions(videoStream: VideoStreamInfo): string[] {
-    return [];
+    return this.getInputThreadOptions();
  }

  getBaseOutputOptions(target: TranscodeTarget, videoStream: VideoStreamInfo, audioStream?: AudioStreamInfo) {
@ -80,11 +84,7 @@ class BaseConfig implements VideoCodecSWConfig {
      options.push(`scale=${this.getScaling(videoStream)}`);
    }

-    if (this.shouldToneMap(videoStream)) {
-      options.push(...this.getToneMapping());
-    }
-    options.push('format=yuv420p');
-
+    options.push(...this.getToneMapping(videoStream), 'format=yuv420p');
    return options;
  }

@ -112,7 +112,11 @@ class BaseConfig implements VideoCodecSWConfig {
    }
  }

-  getThreadOptions(): Array<string> {
+  getInputThreadOptions(): Array<string> {
+    return [];
+  }
+
+  getOutputThreadOptions(): Array<string> {
    if (this.config.threads <= 0) {
      return [];
    }
@ -218,7 +222,11 @@ class BaseConfig implements VideoCodecSWConfig {
    }
  }

-  getToneMapping() {
+  getToneMapping(videoStream: VideoStreamInfo) {
+    if (!this.shouldToneMap(videoStream)) {
+      return [];
+    }
+
    const colors = this.getColors();

    return [
@ -348,8 +356,8 @@ export class ThumbnailConfig extends BaseConfig {
 }

 export class H264Config extends BaseConfig {
-  getThreadOptions() {
-    const options = super.getThreadOptions();
+  getOutputThreadOptions() {
+    const options = super.getOutputThreadOptions();
    if (this.config.threads === 1) {
      options.push('-x264-params frame-threads=1:pools=none');
    }
@ -359,8 +367,8 @@ export class H264Config extends BaseConfig {
 }

 export class HEVCConfig extends BaseConfig {
-  getThreadOptions() {
-    const options = super.getThreadOptions();
+  getOutputThreadOptions() {
+    const options = super.getOutputThreadOptions();
    if (this.config.threads === 1) {
      options.push('-x265-params frame-threads=1:pools=none');
    }
@ -391,8 +399,8 @@ export class VP9Config extends BaseConfig {
    return [`-${this.useCQP() ? 'q:v' : 'crf'} ${this.config.crf}`, `-b:v ${bitrates.max}${bitrates.unit}`];
  }

-  getThreadOptions() {
-    return ['-row-mt 1', ...super.getThreadOptions()];
+  getOutputThreadOptions() {
+    return ['-row-mt 1', ...super.getOutputThreadOptions()];
  }

  eligibleForTwoPass() {
@ -425,7 +433,7 @@ export class AV1Config extends BaseConfig {
    return options;
  }

-  getThreadOptions() {
+  getOutputThreadOptions() {
    return []; // Already set above with svtav1-params
  }

@ -434,7 +442,7 @@ export class AV1Config extends BaseConfig {
  }
 }

-export class NVENCConfig extends BaseHWConfig {
+export class NvencSwDecodeConfig extends BaseHWConfig {
  getSupportedCodecs() {
    return [VideoCodec.H264, VideoCodec.HEVC, VideoCodec.AV1];
  }
@ -462,7 +470,7 @@ export class NVENCConfig extends BaseHWConfig {
  }

  getFilterOptions(videoStream: VideoStreamInfo) {
-    const options = this.shouldToneMap(videoStream) ? this.getToneMapping() : [];
+    const options = this.getToneMapping(videoStream);
    options.push('format=nv12', 'hwupload_cuda');
    if (this.shouldScale(videoStream)) {
      options.push(`scale_cuda=${this.getScaling(videoStream)}`);
@ -513,6 +521,52 @@ export class NVENCConfig extends BaseHWConfig {
  }
 }

+export class NvencHwDecodeConfig extends NvencSwDecodeConfig {
+  getBaseInputOptions() {
+    return ['-hwaccel cuda', '-hwaccel_output_format cuda', '-noautorotate', ...this.getInputThreadOptions()];
+  }
+
+  getFilterOptions(videoStream: VideoStreamInfo) {
+    const options = [];
+    if (this.shouldScale(videoStream)) {
+      options.push(`scale_cuda=${this.getScaling(videoStream)}`);
+    }
+    options.push(...this.getToneMapping(videoStream));
+    if (options.length > 0) {
+      options[options.length - 1] += ':format=nv12';
+    } else {
+      options.push('format=nv12');
+    }
+    return options;
+  }
+
+  getToneMapping(videoStream: VideoStreamInfo) {
+    if (!this.shouldToneMap(videoStream)) {
+      return [];
+    }
+
+    const colors = this.getColors();
+    const tonemapOptions = [
+      'desat=0',
+      `matrix=${colors.matrix}`,
+      `primaries=${colors.primaries}`,
+      'range=pc',
+      `tonemap=${this.config.tonemap}`,
+      `transfer=${colors.transfer}`,
+    ];
+
+    return [`tonemap_cuda=${tonemapOptions.join(':')}`];
+  }
+
+  getInputThreadOptions() {
+    return [`-threads ${this.config.threads <= 0 ? 1 : this.config.threads}`];
+  }
+
+  getOutputThreadOptions() {
+    return [];
+  }
+}
+
 export class QSVConfig extends BaseHWConfig {
  getBaseInputOptions() {
    if (this.devices.length === 0) {
@ -538,7 +592,7 @@ export class QSVConfig extends BaseHWConfig {
  }

  getFilterOptions(videoStream: VideoStreamInfo) {
-    const options = this.shouldToneMap(videoStream) ? this.getToneMapping() : [];
+    const options = this.getToneMapping(videoStream);
    options.push('format=nv12', 'hwupload=extra_hw_frames=64');
    if (this.shouldScale(videoStream)) {
      options.push(`scale_qsv=${this.getScaling(videoStream)}`);
@ -604,7 +658,7 @@ export class VAAPIConfig extends BaseHWConfig {
  }

  getFilterOptions(videoStream: VideoStreamInfo) {
-    const options = this.shouldToneMap(videoStream) ? this.getToneMapping() : [];
+    const options = this.getToneMapping(videoStream);
    options.push('format=nv12', 'hwupload');
    if (this.shouldScale(videoStream)) {
      options.push(`scale_vaapi=${this.getScaling(videoStream)}`);
@ -656,47 +710,22 @@ export class VAAPIConfig extends BaseHWConfig {
  }
 }

-export class RKMPPConfig extends BaseHWConfig {
-  private hasOpenCL: boolean;
-
+export class RkmppSwDecodeConfig extends BaseHWConfig {
  constructor(
    protected config: SystemConfigFFmpegDto,
    devices: string[] = [],
-    hasOpenCL: boolean = false,
  ) {
    super(config, devices);
-    this.hasOpenCL = hasOpenCL;
  }

  eligibleForTwoPass(): boolean {
    return false;
  }

-  getBaseInputOptions(videoStream: VideoStreamInfo) {
+  getBaseInputOptions(): string[] {
    if (this.devices.length === 0) {
      throw new Error('No RKMPP device found');
    }
-    return this.shouldToneMap(videoStream) && !this.hasOpenCL
-      ? [] // disable hardware decoding & filters
-      : ['-hwaccel rkmpp', '-hwaccel_output_format drm_prime', '-afbc rga'];
-  }
-
-  getFilterOptions(videoStream: VideoStreamInfo) {
-    if (this.shouldToneMap(videoStream)) {
-      if (!this.hasOpenCL) {
-        return super.getFilterOptions(videoStream);
-      }
-      const colors = this.getColors();
-      return [
-        `scale_rkrga=${this.getScaling(videoStream)}:format=p010:afbc=1`,
-        'hwmap=derive_device=opencl:mode=read',
-        `tonemap_opencl=format=nv12:r=pc:p=${colors.primaries}:t=${colors.transfer}:m=${colors.matrix}:tonemap=${this.config.tonemap}:desat=0`,
-        'hwmap=derive_device=rkmpp:mode=write:reverse=1',
-        'format=drm_prime',
-      ];
-    } else if (this.shouldScale(videoStream)) {
-      return [`scale_rkrga=${this.getScaling(videoStream)}:format=nv12:afbc=1`];
-    }
    return [];
  }

@ -734,3 +763,29 @@ export class RKMPPConfig extends BaseHWConfig {
    return `${this.config.targetVideoCodec}_rkmpp`;
  }
 }
+
+export class RkmppHwDecodeConfig extends RkmppSwDecodeConfig {
+  getBaseInputOptions() {
+    if (this.devices.length === 0) {
+      throw new Error('No RKMPP device found');
+    }
+
+    return ['-hwaccel rkmpp', '-hwaccel_output_format drm_prime', '-afbc rga'];
+  }
+
+  getFilterOptions(videoStream: VideoStreamInfo) {
+    if (this.shouldToneMap(videoStream)) {
+      const colors = this.getColors();
+      return [
+        `scale_rkrga=${this.getScaling(videoStream)}:format=p010:afbc=1`,
+        'hwmap=derive_device=opencl:mode=read',
+        `tonemap_opencl=format=nv12:r=pc:p=${colors.primaries}:t=${colors.transfer}:m=${colors.matrix}:tonemap=${this.config.tonemap}:desat=0`,
+        'hwmap=derive_device=rkmpp:mode=write:reverse=1',
+        'format=drm_prime',
+      ];
+    } else if (this.shouldScale(videoStream)) {
+      return [`scale_rkrga=${this.getScaling(videoStream)}:format=nv12:afbc=1`];
+    }
+    return [];
+  }
+}
--- a/web/src/lib/components/admin-page/settings/ffmpeg/ffmpeg-settings.svelte
+++ b/web/src/lib/components/admin-page/settings/ffmpeg/ffmpeg-settings.svelte
@ -276,6 +276,15 @@
              isEdited={config.ffmpeg.accel !== savedConfig.ffmpeg.accel}
            />

+            <SettingSwitch
+              id="hardware-decoding"
+              title="HARDWARE DECODING"
+              {disabled}
+              subtitle="Applies only to NVENC and RKMPP. Enables end-to-end acceleration instead of only accelerating encoding. May not work on all videos."
+              bind:checked={config.ffmpeg.accelDecode}
+              isEdited={config.ffmpeg.accelDecode !== savedConfig.ffmpeg.accelDecode}
+            />
+
            <SettingSelect
              label="CONSTANT QUALITY MODE"
              desc="ICQ is better than CQP, but some hardware acceleration devices do not support this mode. Setting this option will prefer the specified mode when using quality-based encoding. Ignored by NVENC as it does not support ICQ."
@ -297,6 +306,7 @@
              bind:checked={config.ffmpeg.temporalAQ}
              isEdited={config.ffmpeg.temporalAQ !== savedConfig.ffmpeg.temporalAQ}
            />
+
            <SettingInputField
              inputType={SettingInputFieldType.TEXT}
              label="PREFERRED HARDWARE DEVICE"